Detection of Characteristic Waves in Clinical Visual Evoked Potential Signals Based on Deep Learning

Abstract

Clinical interpretation of visual evoked potential (VEP) waveforms requires doctors to detect characteristic waves based on their own experiences, which results in variety and inaccuracy. This study proposed a method based on deep learning that can accurately detect the characteristic waves of Flash visual evoked potential (FVEP ) and Pattern visual evoked potential (PVEP ). The proposed method can segment characteristic regions using attention U-Net and locate positions based on the features of characteristic waves. Three FVEP datasets and two PVEP datasets containing a total of 9583 FVEP waveforms and 1997 PVEP waveforms collected by two kinds of instruments from two centres were used in this study. To detect N2 and P2 waves in FVEP data collected in hospital 1, our method achieved precisions of 79.62% and 89.94% and recalls of 77.52% and 90.08%. It achieved precisions of 74.04% and 80.81%, recalls of 73.4% and 82.21%, and precisions of 81.67% and 87.88%, recalls of 81.96% and 87.78% in detecting N2 and P2 waves of FVEP data collected on two kinds of instruments in hospital 2. It also achieved precisions of 94.96% and 97.5% and recalls of 94.29% and 96.47% in detecting P100 waves of PVEP data collected using two kinds of instruments in hospital 2. Our method obtained good accuracy and generalization in detecting characteristic waves of FVEP and PVEP collected by different instruments and showed potential for analyzing various visual electrophysiological signals.

Note:
Funding declaration: This study was supported by grants from the National Natural Science Foundation of China(81974138, SL; 82271054, ZL; U20A20363, JH), Scientific and technological projects with combination of medicine and engineering in Xiamen of China(3502Z20224030, SL), Nature Science Foundation of Fujian Province of China(2022J01110650, SL), and National Key Research &Development Program of China (2018YFA0107301, SL).

Conflict of Interests: None

Ethical Approval: The institutional review board of the Southwest Hospital approved this study (No. KY2020053 2020-05-01). The institutional review board of the School of Medicine, Xiamen University approved this study (identifier, XDYX2022004 2022-02-20). Informed consent in the study was exempted due to using only deidentified retrospective records data without any identifying marks

Keywords: Deep learning, Characteristic Waves Detection, FVEP, PVEP